|
||
|
|
|
              |
|
 |
 |
||||||||||||||
|
 |
|
||||||||||||
By William Van Winkle |
||||||||||||||
The industry is always in a state of ebb and flow, of relative doldrums and spuming storms. No surprises there. But the last quarter of 2007 is shaping up to be a real nor’easter as we have long-awaited changes sweeping over the channel. There are major processor launches, new drive technologies, memory format transitions, shifts in priorities for monitor buyers, and much more. Grab your Gore-Tex and galoshes, everyone. It’s time to get wet. |
||||||||||||||
|
Where there’s change, there’s opportunity. There are also false opportunities sitting in your path like chuckholes, waiting to snag the unwary and dump you on your face. We’ve given a lot of thought to all manner of change and opportunity in this pre-holiday market, and we’ve narrowed the field down to eight areas on which you will want to concentrate your attention for mainstream buyers. To be clear, we’re not out to tell you how to snag a mainstream sale. We want to help you upsell that mainstream purchase and add more value and profit to each invoice. So we’re not going to start our list by telling you to sell notebooks. You’d better be doing that already. And we could chime in with the recommendation to sell VBI whitebooks instead of regular notebooks. But you should know that already, too, or at least be familiar with the benefits offered by Intel’s whitebook program. No, we’re more interested in pointing out technology or usage model changes that can impact how you configure that notebook, not to mention the desktops attached to it across the LAN. Everything changes. Here are some ways to make change a positive for your systems and business.
PREPARE FOR PENRYN In case you’re new to the inner world of CPU manufacturing, let’s take a step back. A processor is manufactured in part by taking a wafer of pure silicon, coating it with certain materials that comprise a “photoresist,” and using short wavelength light to create patterns on the wafer (lithography). These patterns, after additional processes, will become features on the wafer, such as transistor gates. The average size of these features generally becomes known as the “process” or “fabrication node” or something similar. Until now, 2007 has been the year of the 65nm node, meaning that the average length of a transistor gate on a current CPU from Intel or AMD has been 65nm. This is no small trick considering that the light used for this lithography has a wavelength of 193nm and/or 248nm and thus must be used in conjunction with very special and expensive lenses and masks. It’s a bit like figuring out how to use Bundt pans to make cupcakes. The advantage of shrinking the semiconductor fab process is that, for a similar price, you can fit more transistors and other features into the same physical space. Because signals have less distance to travel inside of a chip, there is less chance for their energy to “leak” and create heat. Designers can fit in more processing logic, additional megabytes of cache, and so on. While factors such as smart architecture and broad industry support are critical, history has proven that he who shrinks his process more quickly and can extract from it better manufacturing yields will have a leg up on ruling the market. This is why the current race from 65nm to 45nm is so important. AMD showcased its 45nm capabilities in public for the first time last May, but production won’t start until the “Shanghai” revision to today’s “Barcelona” Opteron, expected in the second half of 2008. (Will you be surprised if AMD pushes Shanghai out just in front of Intel’s 45nm Nehalem launch? We won’t.) However, Intel’s 45nm process, code-named Penryn, should begin shipping right around the time you read this. The 45nm fab shrink itself doesn’t offer a radical overhaul in features, although there are a few. Rather, it’s a fine example of Intel’s “tick-tock” manufacturing strategy, in which the company updates its CPU architecture one year (tick) and its fab process the next (tock). Compared to trying to update both sides at once, this approach reduces the headaches that often accompany major technology advances.
“With 45nm, we will be offering outstanding performance and features at outstanding price points,” notes Intel North American channel marketing manager Todd Garrigues. “That’s the real ‘what’s in it for me?’ question with any transition. How can this help our customers make money? Penryn will provide strong reasons to upgrade at very compelling price points, and this will help move SMB customers into new PCs and servers.” Penryn will first arrive in Intel’s Xeon server/workstation processor line, followed in Q1 by the Core 2 desktop line. For those of you who like to track code names, know that today’s dual-core Woodcrest and quad-core Clovertown—-the two 65nm strains within the Xeon 5000 series—-will become tomorrow’s 45nm Wolfdale DP (5300) and Harpertown (5400). On the desktop side, the 65nm, dual-core Conroe and quad-core Kentsfield will become the 45nm Wolfdale and Yorkfield. Because there is no major architecture change in the offing, you don’t need to get too worried about the names. Just watch the core counts and cache sizes. Whereas 65nm core-based chips had 4MB of L2 cache per dual-core die, the feature size reduction found with Penryn now allows for 6MB of L2 in the same space, making 12MB of L2 cache on each quad-core processor. (This ignores future designs, such as today’s Allendale, in which some on-die resources might be disabled and the chip marketed as a value alternative.) You’ll also find Penryn New Instructions, more commonly known as SSE4. This is a set of 54 new instructions baked into Penryn processors that will help accelerate overall system performance, not just multimedia tasks as some previous MMX and SSE instruction sets have done. Also, Penryn-based Xeon systems still use FD-DIMM memory, but the desktop versions will be able to use either DDR2 or, increasingly often, DDR3.
We’re not here to tell you to buy Penryn chips. When the time and pricing are right, you’re going to do that anyway. Instead, we would urge you to adopt selling desktop motherboards based on Intel’s new 3 Series now. The 3 Series currently divides into four main groups—-the G31/33/35, the P35, the Q33/35, and the X38-—and all of them support not only today’s 65nm Core 2 chips but also the coming wave of Penryn parts. We detailed the 3 Series in last month’s edition of RAM (issue 70), so please hit our Web site (www.reselleradvocate.com) for a refresher if you need one. Suffice it to say here that benefits of the new chipsets include improved integrated graphics, integrated dual-display outputs, new vPro support, front-side bus speeds up to 1333 MHz, DDR3 compatibility, and first-ever support for PCI Express 2.0. The step up to a 3 Series-based motherboard is only a few (if any) dollars more than prior-generation 945/965/975 chipset boards. This is not a good place to skimp. Even if your customer never gets around to upgrading from his 65nm CPU to a 45nm variant, at least you’ve shown that you’re interested in him getting the most longevity and feature quality possible out of his investment. TEXAN INSURRECTION If Intel’s message for the next few months is that the world should revolve around 45nm, then AMD has a rebellious message of its own: Buy unified quad-core. Because while Intel is blowing a large-caliber hole in its profit margins in order to drive quad-core into the mainstream, keep in mind that these quad-core chips are actually multi-chip packages (MCPs). In this case, that means Intel took two of its dual-core dies and “glued” (AMD’s favored term for it) them into a single processor package. For a core on the first die to query a core or cache on the second die, a signal must exit the first die, travel out of the CPU, down the front-side bus into the northbridge, then come back up through the FSB, back to the CPU, and into the second die-—then make the same trip in reverse to provide the query’s results. Not even Intel argues that this is an inefficient architecture compared to AMD’s HyperTransport direct link technology, which allows one processor to speak directly with another and bypass the northbridge.
HyperTransport combined with an integrated memory controller in each CPU is what allowed Opteron to stomp Xeon in the first half of this decade. This is also why you’ll see Intel mimic AMD’s design with its own QuickPath Architecture (formerly called Common System Interconnect, or CSI) and an integrated memory controller in next year’s Nehalem generation. The great R&D minds at AMD in Austin not only kept Hyper-Transport but moved to one-up Intel by being first to market this quarter with a unified quad-core processor. This refers to a design in which four processor cores all share a common set of cache and arbitration resources. Not only can processors talk directly to one another, but now the cores within each processor also have equal, direct access amongst themselves. This is a more elegant architecture, and the sum result is a significant performance improvement in many areas, especially applications that lean heavily on memory and floating point functions. This is not a blank-check endorsement of AMD’s quad-core products in either the consumer or commercial space. An old horse can still beat a young stallion by leveraging brute strength, longer experience, and other factors. There will be situations in which AMD has the superior product for a given task set, and sometimes Intel will be the better solution in a given environment. Early reviews and data comparing AMD’s Barcelona quad-core Opteron and Intel’s quad-core Xeon bear this out. Whether Penryn and the move to having 6MB of L2 cache per Intel dual-core die tilt the scales in Intel’s favor remains to be seen. Our guess is that there will still be a market need for both vendors’ products.
AMD’s native quad-core design comes with other benefits, as well. On the server side, Barcelona still pairs up with DDR2 modules rather than the considerably more power-hungry FB-DIMMs employed by Intel. In fact, there is so much difference in wattage between these two formats that AMD will usually emerge as the more energy-efficient platform when measuring draw from the wall rather than the CPU socket. Power efficiency is probably AMD’s leading quad-core message once you get past the whole unified architecture bit. The new design can set power utilization for each core depending on its load rather than applying one draw for all cores based on the one doing the most work. Large areas in each core can be powered down if they’re not being used, and even tiny transistor groups can go dark, allowing for more dynamic cooling throughout the chip.
But these are chip features, and we’re supposed to be telling you about future-proofing. So here’s the big point on AMD: One of the best new power features in the new quad-core architecture is called Dual Dynamic Power Management. It entails having a split power plane on the motherboard able to route one power stream to the CPU’s cores and another stream to the CPU’s memory controller. This allows for lower overall power draw and a 3% to 10% boost in performance. Obviously, these two points are central to the appeal of AMD’s new architecture. Here’s the catch: While it’s true that the new quad chips will run in last season’s AMD-based motherboards (AM2 and Socket F), those old boards do not support Dual Dynamic Power Management. Only new boards featuring socket AM2+ or the revised Socket F will support the new split-plane functionality. So we would encourage you not to pitch the new AMD quad-core chips as a backward-compatible upgrade solution to customers. Instead, if you’re out to future-proof your clients, we suggest pushing split-plane boards before any discussion of specific processors. Previous-generation AMD chips work just fine on the new motherboards, and customers can preserve their old CPU investment while waiting for prices to come down on the new quad-core parts, most of which haven’t reached the market yet anyway. On new-build PCs, be careful to use the new motherboards, even if they cost a few dollars more. In fact, you can turn this into a positive for your store by asking shoppers if the other bids they’re getting use split plane-enabled boards. If the customer or, even better, the other vendors don’t know, use your knowledge to demonstrate how you take care to maximize the buyer’s investment for both today and tomorrow. ...more |
||||||||||||||
 Back to top
|
||||||||||||||
Copyright © 2007 RAM Magazine. All rights reserved.
Do not duplicate or redistribute in any form. |
||||||||||||||
